LADar: an LLM-based Tool to Detect Web Accessibility Violations

Abstract

Web accessibility is essential to ensure that websites are designed without barriers for people with disabilities. Despite the long-standing standards provided by the Web Content Accessibility Guidelines (WCAG), many web pages remain inaccessible even today. Automated tools such as axe-core and Lighthouse confirm this by identifying accessibility gaps through static analysis of the source code. However, these industry-standard tools often fail to detect violations that require real browser interaction, such as keyboard traps or focus order inconsistencies. In this study, we present LADar, an accessibility auditor that leverages Agentic AI to bridge the gap between static analysis and real browser interaction. By combining rule-based checks with LLM-based reasoning, LADar dynamically plans and executes interaction-driven tests on the rendered page to detect a broader range of WCAG violations. We conducted a preliminary evaluation of the LADar prototype using W3C test pages, additional custom pages based on WCAG failure techniques, and real-world web pages, comparing its performance against axe-core and Lighthouse. Our findings show that static analysis alone is insufficient; however, by integrating real browser interaction and LLM-based reasoning, LADar successfully detects critical violations that traditional automated tools consistently overlook.